Based On Artificial Neural Network Research Machines Olfactory System

Smell simulation technology is a kind of technology which studied how to imitate a biological smell. In early 1960s smell simulator already appeared, but the technology developed rapidly in recent 10 years, and it is widely used in the food analysis, the quality identification of flavor, the environmental detection and the medical and health, etc.This paper is starting from the biological olfactory system, analyses the biological basis of the machine olfactory system, sets out the connection between biological olfactory system and machine olfactory system. It introducts each part of the system specifically, such as Gas sensors and Gas sensor array, signal preprocessing, the common method of statistical pattern recognition and the technique of artificial neural network. On the basis of the above, the machine olfactory system, which is composed of gas sensor array and artificial neural network model identification technology, has been designed and realized. And using the olfactory system designed the experiment of the cigarette type identification by cigarette aroma recognition.The hardware core part of the system is the AT89S52 SCM(Single Chip Micyoco). The system is composed of the sensor array, the sampling circuit, the SCM and peripheral expansion circuit, the sensor heating control circuit and the host (PC), etc. Through the gas sensor array, the smell can be convert to the signals which can be identified by computer. Through RS232 standard serial interface chip, the SCM connected with the host, and upload the data to the PC. Host can also control of the heating and sampling of the sensor. It can acquire and display the real-time sampling data, store and process the data.The system software includes the SCM software and the PC application software. The SCM software completes with assembly language, and it can complete control, I/O operation, and communications functions of SCM system. The application software completes with Visual Basic 6.0 and Matlab. VB is used to develop Windows applications rapidly, and the artificial neural network toolbox of Matlab is used to the design, study, training and simulation of artificial neural network. A combination of VB and Matlab improve the system of data processing ability, and more convenient to use.In order to verify the application characteristics of the machine olfactory system, three kings of cigarettes are selected and made qualitative analysis by it. After a certain scale of trainings by different samples, whether the system can correctly identify unknown cigarette samples can be sure. The experiment shows that the system can identify the aroma of different brands of cigarettes and can identify the type of cigarettes.Effects of different pretreatment algorithms for identification system and structure parameters of neural networks identification modules are compared and analyzed based on datas from experiment, which provide a basis for improving the training speed and the recognize ability of the machine olfactory system.